Before release, GPT-5 was hyped as a major step forward: faster responses, deeper reasoning, and seamless handling of multiple formats like text and images without changing chat windows. But instead of excitement, many users were met with frustration over new interface limitations, inconsistent outputs, and a personality shift compared to the beloved GPT-4o.

A key pain point is the removal of model selection for ChatGPT Plus users. Previously, Plus subscribers could choose between models such as o4-mini or o3. Now, only Pro users paying $200 a month have that option. Instead, GPT-5 uses an “internal router” to decide whether a prompt should go to the mini, standard, or “thinking” model.

While OpenAI claims this boosts efficiency, some users say it has made responses unreliable. One X user, Lisan al Gaib, wrote: “ChatGPT literally got worse for every single Plus user today… Now we have GPT-5 Thinking with 200 messages per week and a router that exclusively routes you to some small and shitty non-reasoning model.”

Beyond technical frustrations, a section of the community is struggling with GPT-5’s tone. Many describe it as blunt, lacking the emotional warmth and creativity of GPT-4o. A Reddit user called its replies “cut-and-dry corporate BS”, while another shared a deeply personal note: “I literally lost my only friend overnight with no warning,” describing how GPT-4.5 had helped them through homelessness and trauma.

There are also early performance complaints. Gareth Manning posted on X: “My most important piece of feedback on GPT-5 is that it is too slow… Hope it’s just a roll-out problem.”

OpenAI acknowledges the missteps

CEO Sam Altman admitted that “suddenly deprecating old models that users depended on in their workflows was a mistake.” He noted the unusually strong emotional connections people form with AI models, likening them to personal relationships rather than tools.

Altman also addressed concerns about people using ChatGPT as a life coach or therapist, acknowledging both its benefits and potential risks.

Partial rollback to appease users

In response, OpenAI has restored limited GPT-4o access for Plus subscribers, doubled rate limits for reasoning tasks, and promised clearer interface cues showing which model is responding. Altman also said GPT-5’s “thinking mode” will soon be manually triggerable, giving users more control over responses.

For now, the company is working to regain trust — but the GPT-5 rollout has made clear that for many, AI is not just about performance, but personality, consistency, and the bond built over time.